xref: /freebsd/sys/geom/raid/md_jmicron.c (revision 9a14aa017b21c292740c00ee098195cd46642730)
1 /*-
2  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/endian.h>
33 #include <sys/kernel.h>
34 #include <sys/kobj.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/mutex.h>
39 #include <sys/systm.h>
40 #include <sys/taskqueue.h>
41 #include <geom/geom.h>
42 #include "geom/raid/g_raid.h"
43 #include "g_raid_md_if.h"
44 
45 static MALLOC_DEFINE(M_MD_JMICRON, "md_jmicron_data", "GEOM_RAID JMicron metadata");
46 
47 #define	JMICRON_MAX_DISKS	8
48 #define	JMICRON_MAX_SPARE	2
49 
50 struct jmicron_raid_conf {
51     u_int8_t		signature[2];
52 #define	JMICRON_MAGIC		"JM"
53 
54     u_int16_t		version;
55 #define	JMICRON_VERSION		0x0001
56 
57     u_int16_t		checksum;
58     u_int8_t		filler_1[10];
59     u_int32_t		disk_id;
60     u_int32_t		offset;
61     u_int32_t		disk_sectors_high;
62     u_int16_t		disk_sectors_low;
63     u_int8_t		filler_2[2];
64     u_int8_t		name[16];
65     u_int8_t		type;
66 #define	JMICRON_T_RAID0		0
67 #define	JMICRON_T_RAID1		1
68 #define	JMICRON_T_RAID01	2
69 #define	JMICRON_T_CONCAT	3
70 #define	JMICRON_T_RAID5		5
71 
72     u_int8_t		stripe_shift;
73     u_int16_t		flags;
74 #define	JMICRON_F_READY		0x0001
75 #define	JMICRON_F_BOOTABLE	0x0002
76 #define	JMICRON_F_BADSEC	0x0004
77 #define	JMICRON_F_ACTIVE	0x0010
78 #define	JMICRON_F_UNSYNC	0x0020
79 #define	JMICRON_F_NEWEST	0x0040
80 
81     u_int8_t		filler_3[4];
82     u_int32_t		spare[JMICRON_MAX_SPARE];
83     u_int32_t		disks[JMICRON_MAX_DISKS];
84 #define	JMICRON_DISK_MASK	0xFFFFFFF0
85 #define	JMICRON_SEG_MASK	0x0000000F
86     u_int8_t		filler_4[32];
87     u_int8_t		filler_5[384];
88 };
89 
90 struct g_raid_md_jmicron_perdisk {
91 	struct jmicron_raid_conf	*pd_meta;
92 	int				 pd_disk_pos;
93 	int				 pd_disk_id;
94 	off_t				 pd_disk_size;
95 };
96 
97 struct g_raid_md_jmicron_object {
98 	struct g_raid_md_object	 mdio_base;
99 	uint32_t		 mdio_config_id;
100 	struct jmicron_raid_conf	*mdio_meta;
101 	struct callout		 mdio_start_co;	/* STARTING state timer. */
102 	int			 mdio_total_disks;
103 	int			 mdio_disks_present;
104 	int			 mdio_started;
105 	int			 mdio_incomplete;
106 	struct root_hold_token	*mdio_rootmount; /* Root mount delay token. */
107 };
108 
109 static g_raid_md_create_t g_raid_md_create_jmicron;
110 static g_raid_md_taste_t g_raid_md_taste_jmicron;
111 static g_raid_md_event_t g_raid_md_event_jmicron;
112 static g_raid_md_ctl_t g_raid_md_ctl_jmicron;
113 static g_raid_md_write_t g_raid_md_write_jmicron;
114 static g_raid_md_fail_disk_t g_raid_md_fail_disk_jmicron;
115 static g_raid_md_free_disk_t g_raid_md_free_disk_jmicron;
116 static g_raid_md_free_t g_raid_md_free_jmicron;
117 
118 static kobj_method_t g_raid_md_jmicron_methods[] = {
119 	KOBJMETHOD(g_raid_md_create,	g_raid_md_create_jmicron),
120 	KOBJMETHOD(g_raid_md_taste,	g_raid_md_taste_jmicron),
121 	KOBJMETHOD(g_raid_md_event,	g_raid_md_event_jmicron),
122 	KOBJMETHOD(g_raid_md_ctl,	g_raid_md_ctl_jmicron),
123 	KOBJMETHOD(g_raid_md_write,	g_raid_md_write_jmicron),
124 	KOBJMETHOD(g_raid_md_fail_disk,	g_raid_md_fail_disk_jmicron),
125 	KOBJMETHOD(g_raid_md_free_disk,	g_raid_md_free_disk_jmicron),
126 	KOBJMETHOD(g_raid_md_free,	g_raid_md_free_jmicron),
127 	{ 0, 0 }
128 };
129 
130 static struct g_raid_md_class g_raid_md_jmicron_class = {
131 	"JMicron",
132 	g_raid_md_jmicron_methods,
133 	sizeof(struct g_raid_md_jmicron_object),
134 	.mdc_priority = 100
135 };
136 
137 static void
138 g_raid_md_jmicron_print(struct jmicron_raid_conf *meta)
139 {
140 	int k;
141 
142 	if (g_raid_debug < 1)
143 		return;
144 
145 	printf("********* ATA JMicron RAID Metadata *********\n");
146 	printf("signature           <%c%c>\n", meta->signature[0], meta->signature[1]);
147 	printf("version             %04x\n", meta->version);
148 	printf("checksum            0x%04x\n", meta->checksum);
149 	printf("disk_id             0x%08x\n", meta->disk_id);
150 	printf("offset              0x%08x\n", meta->offset);
151 	printf("disk_sectors_high   0x%08x\n", meta->disk_sectors_high);
152 	printf("disk_sectors_low    0x%04x\n", meta->disk_sectors_low);
153 	printf("name                <%.16s>\n", meta->name);
154 	printf("type                %d\n", meta->type);
155 	printf("stripe_shift        %d\n", meta->stripe_shift);
156 	printf("flags               %04x\n", meta->flags);
157 	printf("spare              ");
158 	for (k = 0; k < JMICRON_MAX_SPARE; k++)
159 		printf(" 0x%08x", meta->spare[k]);
160 	printf("\n");
161 	printf("disks              ");
162 	for (k = 0; k < JMICRON_MAX_DISKS; k++)
163 		printf(" 0x%08x", meta->disks[k]);
164 	printf("\n");
165 	printf("=================================================\n");
166 }
167 
168 static struct jmicron_raid_conf *
169 jmicron_meta_copy(struct jmicron_raid_conf *meta)
170 {
171 	struct jmicron_raid_conf *nmeta;
172 
173 	nmeta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK);
174 	memcpy(nmeta, meta, sizeof(*meta));
175 	return (nmeta);
176 }
177 
178 static int
179 jmicron_meta_total_disks(struct jmicron_raid_conf *meta)
180 {
181 	int pos;
182 
183 	for (pos = 0; pos < JMICRON_MAX_DISKS; pos++) {
184 		if (meta->disks[pos] == 0)
185 			break;
186 	}
187 	return (pos);
188 }
189 
190 static int
191 jmicron_meta_total_spare(struct jmicron_raid_conf *meta)
192 {
193 	int pos, n;
194 
195 	n = 0;
196 	for (pos = 0; pos < JMICRON_MAX_SPARE; pos++) {
197 		if (meta->spare[pos] != 0)
198 			n++;
199 	}
200 	return (n);
201 }
202 
203 /*
204  * Generate fake Configuration ID based on disk IDs.
205  * Note: it will change after each disk set change.
206  */
207 static uint32_t
208 jmicron_meta_config_id(struct jmicron_raid_conf *meta)
209 {
210 	int pos;
211 	uint32_t config_id;
212 
213 	config_id = 0;
214 	for (pos = 0; pos < JMICRON_MAX_DISKS; pos++)
215 		config_id += meta->disks[pos] << pos;
216 	return (config_id);
217 }
218 
219 static void
220 jmicron_meta_get_name(struct jmicron_raid_conf *meta, char *buf)
221 {
222 	int i;
223 
224 	strncpy(buf, meta->name, 16);
225 	buf[16] = 0;
226 	for (i = 15; i >= 0; i--) {
227 		if (buf[i] > 0x20)
228 			break;
229 		buf[i] = 0;
230 	}
231 }
232 
233 static void
234 jmicron_meta_put_name(struct jmicron_raid_conf *meta, char *buf)
235 {
236 
237 	memset(meta->name, 0x20, 16);
238 	memcpy(meta->name, buf, MIN(strlen(buf), 16));
239 }
240 
241 static int
242 jmicron_meta_find_disk(struct jmicron_raid_conf *meta, uint32_t id)
243 {
244 	int pos;
245 
246 	id &= JMICRON_DISK_MASK;
247 	for (pos = 0; pos < JMICRON_MAX_DISKS; pos++) {
248 		if ((meta->disks[pos] & JMICRON_DISK_MASK) == id)
249 			return (pos);
250 	}
251 	for (pos = 0; pos < JMICRON_MAX_SPARE; pos++) {
252 		if ((meta->spare[pos] & JMICRON_DISK_MASK) == id)
253 			return (-3);
254 	}
255 	return (-1);
256 }
257 
258 static struct jmicron_raid_conf *
259 jmicron_meta_read(struct g_consumer *cp)
260 {
261 	struct g_provider *pp;
262 	struct jmicron_raid_conf *meta;
263 	char *buf;
264 	int error, i;
265 	uint16_t checksum, *ptr;
266 
267 	pp = cp->provider;
268 
269 	/* Read the anchor sector. */
270 	buf = g_read_data(cp,
271 	    pp->mediasize - pp->sectorsize, pp->sectorsize, &error);
272 	if (buf == NULL) {
273 		G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
274 		    pp->name, error);
275 		return (NULL);
276 	}
277 	meta = (struct jmicron_raid_conf *)buf;
278 
279 	/* Check if this is an JMicron RAID struct */
280 	if (strncmp(meta->signature, JMICRON_MAGIC, strlen(JMICRON_MAGIC))) {
281 		G_RAID_DEBUG(1, "JMicron signature check failed on %s", pp->name);
282 		g_free(buf);
283 		return (NULL);
284 	}
285 	meta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK);
286 	memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
287 	g_free(buf);
288 
289 	/* Check metadata checksum. */
290 	for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i < 64; i++)
291 		checksum += *ptr++;
292 	if (checksum != 0) {
293 		G_RAID_DEBUG(1, "JMicron checksum check failed on %s", pp->name);
294 		free(meta, M_MD_JMICRON);
295 		return (NULL);
296 	}
297 
298 	return (meta);
299 }
300 
301 static int
302 jmicron_meta_write(struct g_consumer *cp, struct jmicron_raid_conf *meta)
303 {
304 	struct g_provider *pp;
305 	char *buf;
306 	int error, i;
307 	uint16_t checksum, *ptr;
308 
309 	pp = cp->provider;
310 
311 	/* Recalculate checksum for case if metadata were changed. */
312 	meta->checksum = 0;
313 	for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i < 64; i++)
314 		checksum += *ptr++;
315 	meta->checksum -= checksum;
316 
317 	/* Create and fill buffer. */
318 	buf = malloc(pp->sectorsize, M_MD_JMICRON, M_WAITOK | M_ZERO);
319 	memcpy(buf, meta, sizeof(*meta));
320 
321 	error = g_write_data(cp,
322 	    pp->mediasize - pp->sectorsize, buf, pp->sectorsize);
323 	if (error != 0) {
324 		G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
325 		    pp->name, error);
326 	}
327 
328 	free(buf, M_MD_JMICRON);
329 	return (error);
330 }
331 
332 static int
333 jmicron_meta_erase(struct g_consumer *cp)
334 {
335 	struct g_provider *pp;
336 	char *buf;
337 	int error;
338 
339 	pp = cp->provider;
340 	buf = malloc(pp->sectorsize, M_MD_JMICRON, M_WAITOK | M_ZERO);
341 	error = g_write_data(cp,
342 	    pp->mediasize - pp->sectorsize, buf, pp->sectorsize);
343 	if (error != 0) {
344 		G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
345 		    pp->name, error);
346 	}
347 	free(buf, M_MD_JMICRON);
348 	return (error);
349 }
350 
351 static struct g_raid_disk *
352 g_raid_md_jmicron_get_disk(struct g_raid_softc *sc, int id)
353 {
354 	struct g_raid_disk	*disk;
355 	struct g_raid_md_jmicron_perdisk *pd;
356 
357 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
358 		pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
359 		if (pd->pd_disk_pos == id)
360 			break;
361 	}
362 	return (disk);
363 }
364 
365 static int
366 g_raid_md_jmicron_supported(int level, int qual, int disks, int force)
367 {
368 
369 	if (disks > 8)
370 		return (0);
371 	switch (level) {
372 	case G_RAID_VOLUME_RL_RAID0:
373 		if (disks < 1)
374 			return (0);
375 		if (!force && (disks < 2 || disks > 6))
376 			return (0);
377 		break;
378 	case G_RAID_VOLUME_RL_RAID1:
379 		if (disks < 1)
380 			return (0);
381 		if (!force && (disks != 2))
382 			return (0);
383 		break;
384 	case G_RAID_VOLUME_RL_RAID1E:
385 		if (disks < 2)
386 			return (0);
387 		if (!force && (disks != 4))
388 			return (0);
389 		break;
390 	case G_RAID_VOLUME_RL_SINGLE:
391 		if (disks != 1)
392 			return (0);
393 		if (!force)
394 			return (0);
395 		break;
396 	case G_RAID_VOLUME_RL_CONCAT:
397 		if (disks < 2)
398 			return (0);
399 		break;
400 	case G_RAID_VOLUME_RL_RAID5:
401 		if (disks < 3)
402 			return (0);
403 		if (!force)
404 			return (0);
405 		break;
406 	default:
407 		return (0);
408 	}
409 	if (qual != G_RAID_VOLUME_RLQ_NONE)
410 		return (0);
411 	return (1);
412 }
413 
414 static int
415 g_raid_md_jmicron_start_disk(struct g_raid_disk *disk)
416 {
417 	struct g_raid_softc *sc;
418 	struct g_raid_subdisk *sd, *tmpsd;
419 	struct g_raid_disk *olddisk, *tmpdisk;
420 	struct g_raid_md_object *md;
421 	struct g_raid_md_jmicron_object *mdi;
422 	struct g_raid_md_jmicron_perdisk *pd, *oldpd;
423 	struct jmicron_raid_conf *meta;
424 	int disk_pos, resurrection = 0;
425 
426 	sc = disk->d_softc;
427 	md = sc->sc_md;
428 	mdi = (struct g_raid_md_jmicron_object *)md;
429 	meta = mdi->mdio_meta;
430 	pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
431 	olddisk = NULL;
432 
433 	/* Find disk position in metadata by it's serial. */
434 	if (pd->pd_meta != NULL)
435 		disk_pos = jmicron_meta_find_disk(meta, pd->pd_disk_id);
436 	else
437 		disk_pos = -1;
438 	if (disk_pos < 0) {
439 		G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
440 		/* If we are in the start process, that's all for now. */
441 		if (!mdi->mdio_started)
442 			goto nofit;
443 		/*
444 		 * If we have already started - try to get use of the disk.
445 		 * Try to replace OFFLINE disks first, then FAILED.
446 		 */
447 		TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
448 			if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
449 			    tmpdisk->d_state != G_RAID_DISK_S_FAILED)
450 				continue;
451 			/* Make sure this disk is big enough. */
452 			TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
453 				if (sd->sd_offset + sd->sd_size + 512 >
454 				    pd->pd_disk_size) {
455 					G_RAID_DEBUG1(1, sc,
456 					    "Disk too small (%ju < %ju)",
457 					    pd->pd_disk_size,
458 					    sd->sd_offset + sd->sd_size + 512);
459 					break;
460 				}
461 			}
462 			if (sd != NULL)
463 				continue;
464 			if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
465 				olddisk = tmpdisk;
466 				break;
467 			} else if (olddisk == NULL)
468 				olddisk = tmpdisk;
469 		}
470 		if (olddisk == NULL) {
471 nofit:
472 			if (disk_pos == -3 || pd->pd_disk_pos == -3) {
473 				g_raid_change_disk_state(disk,
474 				    G_RAID_DISK_S_SPARE);
475 				return (1);
476 			} else {
477 				g_raid_change_disk_state(disk,
478 				    G_RAID_DISK_S_STALE);
479 				return (0);
480 			}
481 		}
482 		oldpd = (struct g_raid_md_jmicron_perdisk *)olddisk->d_md_data;
483 		disk_pos = oldpd->pd_disk_pos;
484 		resurrection = 1;
485 	}
486 
487 	if (olddisk == NULL) {
488 		/* Find placeholder by position. */
489 		olddisk = g_raid_md_jmicron_get_disk(sc, disk_pos);
490 		if (olddisk == NULL)
491 			panic("No disk at position %d!", disk_pos);
492 		if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
493 			G_RAID_DEBUG1(1, sc, "More then one disk for pos %d",
494 			    disk_pos);
495 			g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
496 			return (0);
497 		}
498 		oldpd = (struct g_raid_md_jmicron_perdisk *)olddisk->d_md_data;
499 	}
500 
501 	/* Replace failed disk or placeholder with new disk. */
502 	TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
503 		TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
504 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
505 		sd->sd_disk = disk;
506 	}
507 	oldpd->pd_disk_pos = -2;
508 	pd->pd_disk_pos = disk_pos;
509 	/* Update global metadata just in case. */
510 	meta->disks[disk_pos] = pd->pd_disk_id;
511 
512 	/* If it was placeholder -- destroy it. */
513 	if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
514 		g_raid_destroy_disk(olddisk);
515 	} else {
516 		/* Otherwise, make it STALE_FAILED. */
517 		g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
518 	}
519 
520 	/* Welcome the new disk. */
521 	g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
522 	TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
523 
524 		/*
525 		 * Different disks may have different sizes/offsets,
526 		 * especially in concat mode. Update.
527 		 */
528 		if (!resurrection) {
529 			sd->sd_offset =
530 			    (off_t)pd->pd_meta->offset * 16 * 512; //ZZZ
531 			sd->sd_size =
532 			    (((off_t)pd->pd_meta->disk_sectors_high << 16) +
533 			      pd->pd_meta->disk_sectors_low) * 512;
534 		}
535 
536 		if (resurrection) {
537 			/* Stale disk, almost same as new. */
538 			g_raid_change_subdisk_state(sd,
539 			    G_RAID_SUBDISK_S_NEW);
540 		} else if ((meta->flags & JMICRON_F_BADSEC) != 0 &&
541 		    (pd->pd_meta->flags & JMICRON_F_BADSEC) == 0) {
542 			/* Cold-inserted or rebuilding disk. */
543 			g_raid_change_subdisk_state(sd,
544 			    G_RAID_SUBDISK_S_NEW);
545 		} else if (pd->pd_meta->flags & JMICRON_F_UNSYNC) {
546 			/* Dirty or resyncing disk.. */
547 			g_raid_change_subdisk_state(sd,
548 			    G_RAID_SUBDISK_S_STALE);
549 		} else {
550 			/* Up to date disk. */
551 			g_raid_change_subdisk_state(sd,
552 			    G_RAID_SUBDISK_S_ACTIVE);
553 		}
554 		g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
555 		    G_RAID_EVENT_SUBDISK);
556 	}
557 
558 	/* Update status of our need for spare. */
559 	if (mdi->mdio_started) {
560 		mdi->mdio_incomplete =
561 		    (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
562 		     mdi->mdio_total_disks);
563 	}
564 
565 	return (resurrection);
566 }
567 
568 static void
569 g_disk_md_jmicron_retaste(void *arg, int pending)
570 {
571 
572 	G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
573 	g_retaste(&g_raid_class);
574 	free(arg, M_MD_JMICRON);
575 }
576 
577 static void
578 g_raid_md_jmicron_refill(struct g_raid_softc *sc)
579 {
580 	struct g_raid_md_object *md;
581 	struct g_raid_md_jmicron_object *mdi;
582 	struct g_raid_disk *disk;
583 	struct task *task;
584 	int update, na;
585 
586 	md = sc->sc_md;
587 	mdi = (struct g_raid_md_jmicron_object *)md;
588 	update = 0;
589 	do {
590 		/* Make sure we miss anything. */
591 		na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
592 		if (na == mdi->mdio_total_disks)
593 			break;
594 
595 		G_RAID_DEBUG1(1, md->mdo_softc,
596 		    "Array is not complete (%d of %d), "
597 		    "trying to refill.", na, mdi->mdio_total_disks);
598 
599 		/* Try to get use some of STALE disks. */
600 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
601 			if (disk->d_state == G_RAID_DISK_S_STALE) {
602 				update += g_raid_md_jmicron_start_disk(disk);
603 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
604 					break;
605 			}
606 		}
607 		if (disk != NULL)
608 			continue;
609 
610 		/* Try to get use some of SPARE disks. */
611 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
612 			if (disk->d_state == G_RAID_DISK_S_SPARE) {
613 				update += g_raid_md_jmicron_start_disk(disk);
614 				if (disk->d_state == G_RAID_DISK_S_ACTIVE)
615 					break;
616 			}
617 		}
618 	} while (disk != NULL);
619 
620 	/* Write new metadata if we changed something. */
621 	if (update)
622 		g_raid_md_write_jmicron(md, NULL, NULL, NULL);
623 
624 	/* Update status of our need for spare. */
625 	mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
626 	    mdi->mdio_total_disks);
627 
628 	/* Request retaste hoping to find spare. */
629 	if (mdi->mdio_incomplete) {
630 		task = malloc(sizeof(struct task),
631 		    M_MD_JMICRON, M_WAITOK | M_ZERO);
632 		TASK_INIT(task, 0, g_disk_md_jmicron_retaste, task);
633 		taskqueue_enqueue(taskqueue_swi, task);
634 	}
635 }
636 
637 static void
638 g_raid_md_jmicron_start(struct g_raid_softc *sc)
639 {
640 	struct g_raid_md_object *md;
641 	struct g_raid_md_jmicron_object *mdi;
642 	struct g_raid_md_jmicron_perdisk *pd;
643 	struct jmicron_raid_conf *meta;
644 	struct g_raid_volume *vol;
645 	struct g_raid_subdisk *sd;
646 	struct g_raid_disk *disk;
647 	off_t size;
648 	int j, disk_pos;
649 	char buf[17];
650 
651 	md = sc->sc_md;
652 	mdi = (struct g_raid_md_jmicron_object *)md;
653 	meta = mdi->mdio_meta;
654 
655 	/* Create volumes and subdisks. */
656 	jmicron_meta_get_name(meta, buf);
657 	vol = g_raid_create_volume(sc, buf, -1);
658 	size = ((off_t)meta->disk_sectors_high << 16) + meta->disk_sectors_low;
659 	size *= 512; //ZZZ
660 	if (meta->type == JMICRON_T_RAID0) {
661 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
662 		vol->v_mediasize = size * mdi->mdio_total_disks;
663 	} else if (meta->type == JMICRON_T_RAID1) {
664 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
665 		vol->v_mediasize = size;
666 	} else if (meta->type == JMICRON_T_RAID01) {
667 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
668 		vol->v_mediasize = size * mdi->mdio_total_disks / 2;
669 	} else if (meta->type == JMICRON_T_CONCAT) {
670 		if (mdi->mdio_total_disks == 1)
671 			vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
672 		else
673 			vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
674 		vol->v_mediasize = 0;
675 	} else if (meta->type == JMICRON_T_RAID5) {
676 		vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
677 		vol->v_mediasize = size * (mdi->mdio_total_disks - 1);
678 	} else {
679 		vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
680 		vol->v_mediasize = 0;
681 	}
682 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
683 	vol->v_strip_size = 1024 << meta->stripe_shift; //ZZZ
684 	vol->v_disks_count = mdi->mdio_total_disks;
685 	vol->v_sectorsize = 512; //ZZZ
686 	for (j = 0; j < vol->v_disks_count; j++) {
687 		sd = &vol->v_subdisks[j];
688 		sd->sd_offset = (off_t)meta->offset * 16 * 512; //ZZZ
689 		sd->sd_size = size;
690 	}
691 	g_raid_start_volume(vol);
692 
693 	/* Create disk placeholders to store data for later writing. */
694 	for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
695 		pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO);
696 		pd->pd_disk_pos = disk_pos;
697 		pd->pd_disk_id = meta->disks[disk_pos];
698 		disk = g_raid_create_disk(sc);
699 		disk->d_md_data = (void *)pd;
700 		disk->d_state = G_RAID_DISK_S_OFFLINE;
701 		sd = &vol->v_subdisks[disk_pos];
702 		sd->sd_disk = disk;
703 		TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
704 	}
705 
706 	/* Make all disks found till the moment take their places. */
707 	do {
708 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
709 			if (disk->d_state == G_RAID_DISK_S_NONE) {
710 				g_raid_md_jmicron_start_disk(disk);
711 				break;
712 			}
713 		}
714 	} while (disk != NULL);
715 
716 	mdi->mdio_started = 1;
717 	G_RAID_DEBUG1(0, sc, "Array started.");
718 	g_raid_md_write_jmicron(md, NULL, NULL, NULL);
719 
720 	/* Pickup any STALE/SPARE disks to refill array if needed. */
721 	g_raid_md_jmicron_refill(sc);
722 
723 	g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
724 
725 	callout_stop(&mdi->mdio_start_co);
726 	G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
727 	root_mount_rel(mdi->mdio_rootmount);
728 	mdi->mdio_rootmount = NULL;
729 }
730 
731 static void
732 g_raid_md_jmicron_new_disk(struct g_raid_disk *disk)
733 {
734 	struct g_raid_softc *sc;
735 	struct g_raid_md_object *md;
736 	struct g_raid_md_jmicron_object *mdi;
737 	struct jmicron_raid_conf *pdmeta;
738 	struct g_raid_md_jmicron_perdisk *pd;
739 
740 	sc = disk->d_softc;
741 	md = sc->sc_md;
742 	mdi = (struct g_raid_md_jmicron_object *)md;
743 	pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
744 	pdmeta = pd->pd_meta;
745 
746 	if (mdi->mdio_started) {
747 		if (g_raid_md_jmicron_start_disk(disk))
748 			g_raid_md_write_jmicron(md, NULL, NULL, NULL);
749 	} else {
750 		/*
751 		 * If we haven't started yet - update common metadata
752 		 * to get subdisks details, avoiding data from spare disks.
753 		 */
754 		if (mdi->mdio_meta == NULL ||
755 		    jmicron_meta_find_disk(mdi->mdio_meta,
756 		     mdi->mdio_meta->disk_id) == -3) {
757 			if (mdi->mdio_meta != NULL)
758 				free(mdi->mdio_meta, M_MD_JMICRON);
759 			mdi->mdio_meta = jmicron_meta_copy(pdmeta);
760 			mdi->mdio_total_disks = jmicron_meta_total_disks(pdmeta);
761 		}
762 		mdi->mdio_meta->flags |= pdmeta->flags & JMICRON_F_BADSEC;
763 
764 		mdi->mdio_disks_present++;
765 		G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d+%d up)",
766 		    mdi->mdio_disks_present,
767 		    mdi->mdio_total_disks,
768 		    jmicron_meta_total_spare(mdi->mdio_meta));
769 
770 		/* If we collected all needed disks - start array. */
771 		if (mdi->mdio_disks_present == mdi->mdio_total_disks +
772 		    jmicron_meta_total_spare(mdi->mdio_meta))
773 			g_raid_md_jmicron_start(sc);
774 	}
775 }
776 
777 static void
778 g_raid_jmicron_go(void *arg)
779 {
780 	struct g_raid_softc *sc;
781 	struct g_raid_md_object *md;
782 	struct g_raid_md_jmicron_object *mdi;
783 
784 	sc = arg;
785 	md = sc->sc_md;
786 	mdi = (struct g_raid_md_jmicron_object *)md;
787 	if (!mdi->mdio_started) {
788 		G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
789 		g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
790 	}
791 }
792 
793 static int
794 g_raid_md_create_jmicron(struct g_raid_md_object *md, struct g_class *mp,
795     struct g_geom **gp)
796 {
797 	struct g_raid_softc *sc;
798 	struct g_raid_md_jmicron_object *mdi;
799 	char name[16];
800 
801 	mdi = (struct g_raid_md_jmicron_object *)md;
802 	mdi->mdio_config_id = arc4random();
803 	snprintf(name, sizeof(name), "JMicron-%08x", mdi->mdio_config_id);
804 	sc = g_raid_create_node(mp, name, md);
805 	if (sc == NULL)
806 		return (G_RAID_MD_TASTE_FAIL);
807 	md->mdo_softc = sc;
808 	*gp = sc->sc_geom;
809 	return (G_RAID_MD_TASTE_NEW);
810 }
811 
812 static int
813 g_raid_md_taste_jmicron(struct g_raid_md_object *md, struct g_class *mp,
814                               struct g_consumer *cp, struct g_geom **gp)
815 {
816 	struct g_consumer *rcp;
817 	struct g_provider *pp;
818 	struct g_raid_md_jmicron_object *mdi, *mdi1;
819 	struct g_raid_softc *sc;
820 	struct g_raid_disk *disk;
821 	struct jmicron_raid_conf *meta;
822 	struct g_raid_md_jmicron_perdisk *pd;
823 	struct g_geom *geom;
824 	int error, disk_pos, result, spare, len;
825 	char name[16];
826 	uint16_t vendor;
827 
828 	G_RAID_DEBUG(1, "Tasting JMicron on %s", cp->provider->name);
829 	mdi = (struct g_raid_md_jmicron_object *)md;
830 	pp = cp->provider;
831 
832 	/* Read metadata from device. */
833 	meta = NULL;
834 	vendor = 0xffff;
835 	if (g_access(cp, 1, 0, 0) != 0)
836 		return (G_RAID_MD_TASTE_FAIL);
837 	g_topology_unlock();
838 	len = 2;
839 	if (pp->geom->rank == 1)
840 		g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
841 	meta = jmicron_meta_read(cp);
842 	g_topology_lock();
843 	g_access(cp, -1, 0, 0);
844 	if (meta == NULL) {
845 		if (g_raid_aggressive_spare) {
846 			if (vendor == 0x197b) {
847 				G_RAID_DEBUG(1,
848 				    "No JMicron metadata, forcing spare.");
849 				spare = 2;
850 				goto search;
851 			} else {
852 				G_RAID_DEBUG(1,
853 				    "JMicron vendor mismatch 0x%04x != 0x197b",
854 				    vendor);
855 			}
856 		}
857 		return (G_RAID_MD_TASTE_FAIL);
858 	}
859 
860 	/* Check this disk position in obtained metadata. */
861 	disk_pos = jmicron_meta_find_disk(meta, meta->disk_id);
862 	if (disk_pos == -1) {
863 		G_RAID_DEBUG(1, "JMicron disk_id %08x not found",
864 		    meta->disk_id);
865 		goto fail1;
866 	}
867 
868 	/* Metadata valid. Print it. */
869 	g_raid_md_jmicron_print(meta);
870 	G_RAID_DEBUG(1, "JMicron disk position %d", disk_pos);
871 	spare = (disk_pos == -2) ? 1 : 0;
872 
873 search:
874 	/* Search for matching node. */
875 	sc = NULL;
876 	mdi1 = NULL;
877 	LIST_FOREACH(geom, &mp->geom, geom) {
878 		sc = geom->softc;
879 		if (sc == NULL)
880 			continue;
881 		if (sc->sc_stopping != 0)
882 			continue;
883 		if (sc->sc_md->mdo_class != md->mdo_class)
884 			continue;
885 		mdi1 = (struct g_raid_md_jmicron_object *)sc->sc_md;
886 		if (spare == 2) {
887 			if (mdi1->mdio_incomplete)
888 				break;
889 		} else {
890 			if (mdi1->mdio_config_id ==
891 			    jmicron_meta_config_id(meta))
892 				break;
893 		}
894 	}
895 
896 	/* Found matching node. */
897 	if (geom != NULL) {
898 		G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
899 		result = G_RAID_MD_TASTE_EXISTING;
900 
901 	} else if (spare) { /* Not found needy node -- left for later. */
902 		G_RAID_DEBUG(1, "Spare is not needed at this time");
903 		goto fail1;
904 
905 	} else { /* Not found matching node -- create one. */
906 		result = G_RAID_MD_TASTE_NEW;
907 		mdi->mdio_config_id = jmicron_meta_config_id(meta);
908 		snprintf(name, sizeof(name), "JMicron-%08x",
909 		    mdi->mdio_config_id);
910 		sc = g_raid_create_node(mp, name, md);
911 		md->mdo_softc = sc;
912 		geom = sc->sc_geom;
913 		callout_init(&mdi->mdio_start_co, 1);
914 		callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
915 		    g_raid_jmicron_go, sc);
916 		mdi->mdio_rootmount = root_mount_hold("GRAID-JMicron");
917 		G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
918 	}
919 
920 	rcp = g_new_consumer(geom);
921 	g_attach(rcp, pp);
922 	if (g_access(rcp, 1, 1, 1) != 0)
923 		; //goto fail1;
924 
925 	g_topology_unlock();
926 	sx_xlock(&sc->sc_lock);
927 
928 	pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO);
929 	pd->pd_meta = meta;
930 	if (spare == 2) {
931 		pd->pd_disk_pos = -3;
932 		pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK;
933 	} else {
934 		pd->pd_disk_pos = -1;
935 		pd->pd_disk_id = meta->disk_id;
936 	}
937 	pd->pd_disk_size = pp->mediasize;
938 	disk = g_raid_create_disk(sc);
939 	disk->d_md_data = (void *)pd;
940 	disk->d_consumer = rcp;
941 	rcp->private = disk;
942 
943 	/* Read kernel dumping information. */
944 	disk->d_kd.offset = 0;
945 	disk->d_kd.length = OFF_MAX;
946 	len = sizeof(disk->d_kd);
947 	error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd);
948 	if (disk->d_kd.di.dumper == NULL)
949 		G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.",
950 		    rcp->provider->name, error);
951 
952 	g_raid_md_jmicron_new_disk(disk);
953 
954 	sx_xunlock(&sc->sc_lock);
955 	g_topology_lock();
956 	*gp = geom;
957 	return (result);
958 fail1:
959 	free(meta, M_MD_JMICRON);
960 	return (G_RAID_MD_TASTE_FAIL);
961 }
962 
963 static int
964 g_raid_md_event_jmicron(struct g_raid_md_object *md,
965     struct g_raid_disk *disk, u_int event)
966 {
967 	struct g_raid_softc *sc;
968 	struct g_raid_subdisk *sd;
969 	struct g_raid_md_jmicron_object *mdi;
970 	struct g_raid_md_jmicron_perdisk *pd;
971 
972 	sc = md->mdo_softc;
973 	mdi = (struct g_raid_md_jmicron_object *)md;
974 	if (disk == NULL) {
975 		switch (event) {
976 		case G_RAID_NODE_E_START:
977 			if (!mdi->mdio_started)
978 				g_raid_md_jmicron_start(sc);
979 			return (0);
980 		}
981 		return (-1);
982 	}
983 	pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
984 	switch (event) {
985 	case G_RAID_DISK_E_DISCONNECTED:
986 		/* If disk was assigned, just update statuses. */
987 		if (pd->pd_disk_pos >= 0) {
988 			g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
989 			if (disk->d_consumer) {
990 				g_raid_kill_consumer(sc, disk->d_consumer);
991 				disk->d_consumer = NULL;
992 			}
993 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
994 				g_raid_change_subdisk_state(sd,
995 				    G_RAID_SUBDISK_S_NONE);
996 				g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
997 				    G_RAID_EVENT_SUBDISK);
998 			}
999 		} else {
1000 			/* Otherwise -- delete. */
1001 			g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1002 			g_raid_destroy_disk(disk);
1003 		}
1004 
1005 		/* Write updated metadata to all disks. */
1006 		g_raid_md_write_jmicron(md, NULL, NULL, NULL);
1007 
1008 		/* Check if anything left except placeholders. */
1009 		if (g_raid_ndisks(sc, -1) ==
1010 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1011 			g_raid_destroy_node(sc, 0);
1012 		else
1013 			g_raid_md_jmicron_refill(sc);
1014 		return (0);
1015 	}
1016 	return (-2);
1017 }
1018 
1019 static int
1020 g_raid_md_ctl_jmicron(struct g_raid_md_object *md,
1021     struct gctl_req *req)
1022 {
1023 	struct g_raid_softc *sc;
1024 	struct g_raid_volume *vol;
1025 	struct g_raid_subdisk *sd;
1026 	struct g_raid_disk *disk;
1027 	struct g_raid_md_jmicron_object *mdi;
1028 	struct g_raid_md_jmicron_perdisk *pd;
1029 	struct g_consumer *cp;
1030 	struct g_provider *pp;
1031 	char arg[16];
1032 	const char *verb, *volname, *levelname, *diskname;
1033 	int *nargs, *force;
1034 	off_t size, sectorsize, strip;
1035 	intmax_t *sizearg, *striparg;
1036 	int numdisks, i, len, level, qual, update;
1037 	int error;
1038 
1039 	sc = md->mdo_softc;
1040 	mdi = (struct g_raid_md_jmicron_object *)md;
1041 	verb = gctl_get_param(req, "verb", NULL);
1042 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1043 	error = 0;
1044 	if (strcmp(verb, "label") == 0) {
1045 
1046 		if (*nargs < 4) {
1047 			gctl_error(req, "Invalid number of arguments.");
1048 			return (-1);
1049 		}
1050 		volname = gctl_get_asciiparam(req, "arg1");
1051 		if (volname == NULL) {
1052 			gctl_error(req, "No volume name.");
1053 			return (-2);
1054 		}
1055 		levelname = gctl_get_asciiparam(req, "arg2");
1056 		if (levelname == NULL) {
1057 			gctl_error(req, "No RAID level.");
1058 			return (-3);
1059 		}
1060 		if (g_raid_volume_str2level(levelname, &level, &qual)) {
1061 			gctl_error(req, "Unknown RAID level '%s'.", levelname);
1062 			return (-4);
1063 		}
1064 		numdisks = *nargs - 3;
1065 		force = gctl_get_paraml(req, "force", sizeof(*force));
1066 		if (!g_raid_md_jmicron_supported(level, qual, numdisks,
1067 		    force ? *force : 0)) {
1068 			gctl_error(req, "Unsupported RAID level "
1069 			    "(0x%02x/0x%02x), or number of disks (%d).",
1070 			    level, qual, numdisks);
1071 			return (-5);
1072 		}
1073 
1074 		/* Search for disks, connect them and probe. */
1075 		size = 0x7fffffffffffffffllu;
1076 		sectorsize = 0;
1077 		for (i = 0; i < numdisks; i++) {
1078 			snprintf(arg, sizeof(arg), "arg%d", i + 3);
1079 			diskname = gctl_get_asciiparam(req, arg);
1080 			if (diskname == NULL) {
1081 				gctl_error(req, "No disk name (%s).", arg);
1082 				error = -6;
1083 				break;
1084 			}
1085 			if (strcmp(diskname, "NONE") == 0) {
1086 				cp = NULL;
1087 				pp = NULL;
1088 			} else {
1089 				g_topology_lock();
1090 				cp = g_raid_open_consumer(sc, diskname);
1091 				if (cp == NULL) {
1092 					gctl_error(req, "Can't open '%s'.",
1093 					    diskname);
1094 					g_topology_unlock();
1095 					error = -7;
1096 					break;
1097 				}
1098 				pp = cp->provider;
1099 			}
1100 			pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO);
1101 			pd->pd_disk_pos = i;
1102 			pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK;
1103 			disk = g_raid_create_disk(sc);
1104 			disk->d_md_data = (void *)pd;
1105 			disk->d_consumer = cp;
1106 			if (cp == NULL)
1107 				continue;
1108 			cp->private = disk;
1109 			g_topology_unlock();
1110 
1111 			/* Read kernel dumping information. */
1112 			disk->d_kd.offset = 0;
1113 			disk->d_kd.length = OFF_MAX;
1114 			len = sizeof(disk->d_kd);
1115 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1116 			if (disk->d_kd.di.dumper == NULL)
1117 				G_RAID_DEBUG1(2, sc,
1118 				    "Dumping not supported by %s.",
1119 				    cp->provider->name);
1120 
1121 			pd->pd_disk_size = pp->mediasize;
1122 			if (size > pp->mediasize)
1123 				size = pp->mediasize;
1124 			if (sectorsize < pp->sectorsize)
1125 				sectorsize = pp->sectorsize;
1126 		}
1127 		if (error != 0)
1128 			return (error);
1129 
1130 		if (sectorsize <= 0) {
1131 			gctl_error(req, "Can't get sector size.");
1132 			return (-8);
1133 		}
1134 
1135 		/* Reserve space for metadata. */
1136 		size -= sectorsize;
1137 
1138 		/* Handle size argument. */
1139 		len = sizeof(*sizearg);
1140 		sizearg = gctl_get_param(req, "size", &len);
1141 		if (sizearg != NULL && len == sizeof(*sizearg) &&
1142 		    *sizearg > 0) {
1143 			if (*sizearg > size) {
1144 				gctl_error(req, "Size too big %lld > %lld.",
1145 				    (long long)*sizearg, (long long)size);
1146 				return (-9);
1147 			}
1148 			size = *sizearg;
1149 		}
1150 
1151 		/* Handle strip argument. */
1152 		strip = 131072;
1153 		len = sizeof(*striparg);
1154 		striparg = gctl_get_param(req, "strip", &len);
1155 		if (striparg != NULL && len == sizeof(*striparg) &&
1156 		    *striparg > 0) {
1157 			if (*striparg < sectorsize) {
1158 				gctl_error(req, "Strip size too small.");
1159 				return (-10);
1160 			}
1161 			if (*striparg % sectorsize != 0) {
1162 				gctl_error(req, "Incorrect strip size.");
1163 				return (-11);
1164 			}
1165 			if (strip > 65535 * sectorsize) {
1166 				gctl_error(req, "Strip size too big.");
1167 				return (-12);
1168 			}
1169 			strip = *striparg;
1170 		}
1171 
1172 		/* Round size down to strip or sector. */
1173 		if (level == G_RAID_VOLUME_RL_RAID1)
1174 			size -= (size % sectorsize);
1175 		else if (level == G_RAID_VOLUME_RL_RAID1E &&
1176 		    (numdisks & 1) != 0)
1177 			size -= (size % (2 * strip));
1178 		else
1179 			size -= (size % strip);
1180 		if (size <= 0) {
1181 			gctl_error(req, "Size too small.");
1182 			return (-13);
1183 		}
1184 		if (size > 0xffffffffffffllu * sectorsize) {
1185 			gctl_error(req, "Size too big.");
1186 			return (-14);
1187 		}
1188 
1189 		/* We have all we need, create things: volume, ... */
1190 		mdi->mdio_total_disks = numdisks;
1191 		mdi->mdio_started = 1;
1192 		vol = g_raid_create_volume(sc, volname, -1);
1193 		vol->v_md_data = (void *)(intptr_t)0;
1194 		vol->v_raid_level = level;
1195 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
1196 		vol->v_strip_size = strip;
1197 		vol->v_disks_count = numdisks;
1198 		if (level == G_RAID_VOLUME_RL_RAID0 ||
1199 		    level == G_RAID_VOLUME_RL_CONCAT ||
1200 		    level == G_RAID_VOLUME_RL_SINGLE)
1201 			vol->v_mediasize = size * numdisks;
1202 		else if (level == G_RAID_VOLUME_RL_RAID1)
1203 			vol->v_mediasize = size;
1204 		else if (level == G_RAID_VOLUME_RL_RAID5)
1205 			vol->v_mediasize = size * (numdisks - 1);
1206 		else { /* RAID1E */
1207 			vol->v_mediasize = ((size * numdisks) / strip / 2) *
1208 			    strip;
1209 		}
1210 		vol->v_sectorsize = sectorsize;
1211 		g_raid_start_volume(vol);
1212 
1213 		/* , and subdisks. */
1214 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1215 			pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1216 			sd = &vol->v_subdisks[pd->pd_disk_pos];
1217 			sd->sd_disk = disk;
1218 			sd->sd_offset = 0;
1219 			sd->sd_size = size;
1220 			TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1221 			if (sd->sd_disk->d_consumer != NULL) {
1222 				g_raid_change_disk_state(disk,
1223 				    G_RAID_DISK_S_ACTIVE);
1224 				g_raid_change_subdisk_state(sd,
1225 				    G_RAID_SUBDISK_S_ACTIVE);
1226 				g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1227 				    G_RAID_EVENT_SUBDISK);
1228 			} else {
1229 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1230 			}
1231 		}
1232 
1233 		/* Write metadata based on created entities. */
1234 		G_RAID_DEBUG1(0, sc, "Array started.");
1235 		g_raid_md_write_jmicron(md, NULL, NULL, NULL);
1236 
1237 		/* Pickup any STALE/SPARE disks to refill array if needed. */
1238 		g_raid_md_jmicron_refill(sc);
1239 
1240 		g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1241 		    G_RAID_EVENT_VOLUME);
1242 		return (0);
1243 	}
1244 	if (strcmp(verb, "delete") == 0) {
1245 
1246 		/* Check if some volume is still open. */
1247 		force = gctl_get_paraml(req, "force", sizeof(*force));
1248 		if (force != NULL && *force == 0 &&
1249 		    g_raid_nopens(sc) != 0) {
1250 			gctl_error(req, "Some volume is still open.");
1251 			return (-4);
1252 		}
1253 
1254 		TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1255 			if (disk->d_consumer)
1256 				jmicron_meta_erase(disk->d_consumer);
1257 		}
1258 		g_raid_destroy_node(sc, 0);
1259 		return (0);
1260 	}
1261 	if (strcmp(verb, "remove") == 0 ||
1262 	    strcmp(verb, "fail") == 0) {
1263 		if (*nargs < 2) {
1264 			gctl_error(req, "Invalid number of arguments.");
1265 			return (-1);
1266 		}
1267 		for (i = 1; i < *nargs; i++) {
1268 			snprintf(arg, sizeof(arg), "arg%d", i);
1269 			diskname = gctl_get_asciiparam(req, arg);
1270 			if (diskname == NULL) {
1271 				gctl_error(req, "No disk name (%s).", arg);
1272 				error = -2;
1273 				break;
1274 			}
1275 			if (strncmp(diskname, "/dev/", 5) == 0)
1276 				diskname += 5;
1277 
1278 			TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1279 				if (disk->d_consumer != NULL &&
1280 				    disk->d_consumer->provider != NULL &&
1281 				    strcmp(disk->d_consumer->provider->name,
1282 				     diskname) == 0)
1283 					break;
1284 			}
1285 			if (disk == NULL) {
1286 				gctl_error(req, "Disk '%s' not found.",
1287 				    diskname);
1288 				error = -3;
1289 				break;
1290 			}
1291 
1292 			if (strcmp(verb, "fail") == 0) {
1293 				g_raid_md_fail_disk_jmicron(md, NULL, disk);
1294 				continue;
1295 			}
1296 
1297 			pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1298 
1299 			/* Erase metadata on deleting disk. */
1300 			jmicron_meta_erase(disk->d_consumer);
1301 
1302 			/* If disk was assigned, just update statuses. */
1303 			if (pd->pd_disk_pos >= 0) {
1304 				g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1305 				g_raid_kill_consumer(sc, disk->d_consumer);
1306 				disk->d_consumer = NULL;
1307 				TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1308 					g_raid_change_subdisk_state(sd,
1309 					    G_RAID_SUBDISK_S_NONE);
1310 					g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1311 					    G_RAID_EVENT_SUBDISK);
1312 				}
1313 			} else {
1314 				/* Otherwise -- delete. */
1315 				g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1316 				g_raid_destroy_disk(disk);
1317 			}
1318 		}
1319 
1320 		/* Write updated metadata to remaining disks. */
1321 		g_raid_md_write_jmicron(md, NULL, NULL, NULL);
1322 
1323 		/* Check if anything left except placeholders. */
1324 		if (g_raid_ndisks(sc, -1) ==
1325 		    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1326 			g_raid_destroy_node(sc, 0);
1327 		else
1328 			g_raid_md_jmicron_refill(sc);
1329 		return (error);
1330 	}
1331 	if (strcmp(verb, "insert") == 0) {
1332 		if (*nargs < 2) {
1333 			gctl_error(req, "Invalid number of arguments.");
1334 			return (-1);
1335 		}
1336 		update = 0;
1337 		for (i = 1; i < *nargs; i++) {
1338 			/* Get disk name. */
1339 			snprintf(arg, sizeof(arg), "arg%d", i);
1340 			diskname = gctl_get_asciiparam(req, arg);
1341 			if (diskname == NULL) {
1342 				gctl_error(req, "No disk name (%s).", arg);
1343 				error = -3;
1344 				break;
1345 			}
1346 
1347 			/* Try to find provider with specified name. */
1348 			g_topology_lock();
1349 			cp = g_raid_open_consumer(sc, diskname);
1350 			if (cp == NULL) {
1351 				gctl_error(req, "Can't open disk '%s'.",
1352 				    diskname);
1353 				g_topology_unlock();
1354 				error = -4;
1355 				break;
1356 			}
1357 			pp = cp->provider;
1358 
1359 			pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO);
1360 			pd->pd_disk_pos = -3;
1361 			pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK;
1362 			pd->pd_disk_size = pp->mediasize;
1363 
1364 			disk = g_raid_create_disk(sc);
1365 			disk->d_consumer = cp;
1366 			disk->d_md_data = (void *)pd;
1367 			cp->private = disk;
1368 			g_topology_unlock();
1369 
1370 			/* Read kernel dumping information. */
1371 			disk->d_kd.offset = 0;
1372 			disk->d_kd.length = OFF_MAX;
1373 			len = sizeof(disk->d_kd);
1374 			g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
1375 			if (disk->d_kd.di.dumper == NULL)
1376 				G_RAID_DEBUG1(2, sc,
1377 				    "Dumping not supported by %s.",
1378 				    cp->provider->name);
1379 
1380 			/* Welcome the "new" disk. */
1381 			update += g_raid_md_jmicron_start_disk(disk);
1382 			if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1383 			    disk->d_state != G_RAID_DISK_S_SPARE) {
1384 				gctl_error(req, "Disk '%s' doesn't fit.",
1385 				    diskname);
1386 				g_raid_destroy_disk(disk);
1387 				error = -8;
1388 				break;
1389 			}
1390 		}
1391 
1392 		/* Write new metadata if we changed something. */
1393 		if (update)
1394 			g_raid_md_write_jmicron(md, NULL, NULL, NULL);
1395 		return (error);
1396 	}
1397 	gctl_error(req, "Command '%s' is not supported.", verb);
1398 	return (-100);
1399 }
1400 
1401 static int
1402 g_raid_md_write_jmicron(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1403     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1404 {
1405 	struct g_raid_softc *sc;
1406 	struct g_raid_volume *vol;
1407 	struct g_raid_subdisk *sd;
1408 	struct g_raid_disk *disk;
1409 	struct g_raid_md_jmicron_object *mdi;
1410 	struct g_raid_md_jmicron_perdisk *pd;
1411 	struct jmicron_raid_conf *meta;
1412 	int i, spares;
1413 
1414 	sc = md->mdo_softc;
1415 	mdi = (struct g_raid_md_jmicron_object *)md;
1416 
1417 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1418 		return (0);
1419 
1420 	/* There is only one volume. */
1421 	vol = TAILQ_FIRST(&sc->sc_volumes);
1422 
1423 	/* Fill global fields. */
1424 	meta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK | M_ZERO);
1425 	strncpy(meta->signature, JMICRON_MAGIC, 2);
1426 	meta->version = JMICRON_VERSION;
1427 	jmicron_meta_put_name(meta, vol->v_name);
1428 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
1429 		meta->type = JMICRON_T_RAID0;
1430 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1431 		meta->type = JMICRON_T_RAID1;
1432 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1433 		meta->type = JMICRON_T_RAID01;
1434 	else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT ||
1435 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1436 		meta->type = JMICRON_T_CONCAT;
1437 	else
1438 		meta->type = JMICRON_T_RAID5;
1439 	meta->stripe_shift = fls(vol->v_strip_size / 2048);
1440 	meta->flags = JMICRON_F_READY | JMICRON_F_BOOTABLE;
1441 	for (i = 0; i < vol->v_disks_count; i++) {
1442 		sd = &vol->v_subdisks[i];
1443 		if (sd->sd_disk == NULL || sd->sd_disk->d_md_data == NULL)
1444 			meta->disks[i] = 0xffffffff;
1445 		else {
1446 			pd = (struct g_raid_md_jmicron_perdisk *)
1447 			    sd->sd_disk->d_md_data;
1448 			meta->disks[i] = pd->pd_disk_id;
1449 		}
1450 		if (sd->sd_state < G_RAID_SUBDISK_S_STALE)
1451 			meta->flags |= JMICRON_F_BADSEC;
1452 		if (vol->v_dirty)
1453 			meta->flags |= JMICRON_F_UNSYNC;
1454 	}
1455 
1456 	/* Put spares to their slots. */
1457 	spares = 0;
1458 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1459 		pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1460 		if (disk->d_state != G_RAID_DISK_S_SPARE)
1461 			continue;
1462 		meta->spare[spares] = pd->pd_disk_id;
1463 		if (++spares >= 2)
1464 			break;
1465 	}
1466 
1467 	/* We are done. Print meta data and store them to disks. */
1468 	if (mdi->mdio_meta != NULL)
1469 		free(mdi->mdio_meta, M_MD_JMICRON);
1470 	mdi->mdio_meta = meta;
1471 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1472 		pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1473 		if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1474 		    disk->d_state != G_RAID_DISK_S_SPARE)
1475 			continue;
1476 		if (pd->pd_meta != NULL) {
1477 			free(pd->pd_meta, M_MD_JMICRON);
1478 			pd->pd_meta = NULL;
1479 		}
1480 		pd->pd_meta = jmicron_meta_copy(meta);
1481 		pd->pd_meta->disk_id = pd->pd_disk_id;
1482 		if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) {
1483 			pd->pd_meta->offset =
1484 			    (sd->sd_offset / 512) / 16;
1485 			pd->pd_meta->disk_sectors_high =
1486 			    (sd->sd_size / 512) >> 16;
1487 			pd->pd_meta->disk_sectors_low =
1488 			    (sd->sd_size / 512) & 0xffff;
1489 			if (sd->sd_state < G_RAID_SUBDISK_S_STALE)
1490 				pd->pd_meta->flags &= ~JMICRON_F_BADSEC;
1491 			else if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE)
1492 				pd->pd_meta->flags |= JMICRON_F_UNSYNC;
1493 		}
1494 		G_RAID_DEBUG(1, "Writing JMicron metadata to %s",
1495 		    g_raid_get_diskname(disk));
1496 		g_raid_md_jmicron_print(pd->pd_meta);
1497 		jmicron_meta_write(disk->d_consumer, pd->pd_meta);
1498 	}
1499 	return (0);
1500 }
1501 
1502 static int
1503 g_raid_md_fail_disk_jmicron(struct g_raid_md_object *md,
1504     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1505 {
1506 	struct g_raid_softc *sc;
1507 	struct g_raid_md_jmicron_perdisk *pd;
1508 	struct g_raid_subdisk *sd;
1509 
1510 	sc = md->mdo_softc;
1511 	pd = (struct g_raid_md_jmicron_perdisk *)tdisk->d_md_data;
1512 
1513 	/* We can't fail disk that is not a part of array now. */
1514 	if (pd->pd_disk_pos < 0)
1515 		return (-1);
1516 
1517 	if (tdisk->d_consumer)
1518 		jmicron_meta_erase(tdisk->d_consumer);
1519 
1520 	/* Change states. */
1521 	g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1522 	TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1523 		g_raid_change_subdisk_state(sd,
1524 		    G_RAID_SUBDISK_S_FAILED);
1525 		g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1526 		    G_RAID_EVENT_SUBDISK);
1527 	}
1528 
1529 	/* Write updated metadata to remaining disks. */
1530 	g_raid_md_write_jmicron(md, NULL, NULL, tdisk);
1531 
1532 	/* Check if anything left except placeholders. */
1533 	if (g_raid_ndisks(sc, -1) ==
1534 	    g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1535 		g_raid_destroy_node(sc, 0);
1536 	else
1537 		g_raid_md_jmicron_refill(sc);
1538 	return (0);
1539 }
1540 
1541 static int
1542 g_raid_md_free_disk_jmicron(struct g_raid_md_object *md,
1543     struct g_raid_disk *disk)
1544 {
1545 	struct g_raid_md_jmicron_perdisk *pd;
1546 
1547 	pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data;
1548 	if (pd->pd_meta != NULL) {
1549 		free(pd->pd_meta, M_MD_JMICRON);
1550 		pd->pd_meta = NULL;
1551 	}
1552 	free(pd, M_MD_JMICRON);
1553 	disk->d_md_data = NULL;
1554 	return (0);
1555 }
1556 
1557 static int
1558 g_raid_md_free_jmicron(struct g_raid_md_object *md)
1559 {
1560 	struct g_raid_md_jmicron_object *mdi;
1561 
1562 	mdi = (struct g_raid_md_jmicron_object *)md;
1563 	if (!mdi->mdio_started) {
1564 		mdi->mdio_started = 0;
1565 		callout_stop(&mdi->mdio_start_co);
1566 		G_RAID_DEBUG1(1, md->mdo_softc,
1567 		    "root_mount_rel %p", mdi->mdio_rootmount);
1568 		root_mount_rel(mdi->mdio_rootmount);
1569 		mdi->mdio_rootmount = NULL;
1570 	}
1571 	if (mdi->mdio_meta != NULL) {
1572 		free(mdi->mdio_meta, M_MD_JMICRON);
1573 		mdi->mdio_meta = NULL;
1574 	}
1575 	return (0);
1576 }
1577 
1578 G_RAID_MD_DECLARE(g_raid_md_jmicron);
1579